NHL Goalie and Skater Heat Maps
Sasank Vishnubhatla
January 10, 2019
Last Update: 2019-01-12 13:04:22
GitHub Repository: sv4u/goalie-and-skater-heat-maps
Introduction
In the NHL, players have their sweet-spots. For skaters, it can be the top of a certain circle, or right in-front of the net. For goalies, it could be where their vision is best and where they have the best angle to cut down a shot. All players have these sweet-spots, but it is difficult to analytically say where they are. By using shot location data, we can determine these locations and create models show where goalies and skaters need improvement and where they succeed.
Before we jump in, let’s clean up our R environment and also load in some libraries we will be using.
rm(list = ls())
library(purrr)
library(ggplot2)Data Formatting
To start, we need to read in our data. Our data is formatted nicely in CSV format. We have data from the 2016-2017 season, 2017-2018 season, and 2018-2019 season (up to 1/7/19). This data was downloaded from MoneyPuck. Let’s first start by loading in all three seasons of data:
data.2016 = read.csv("data/2016.csv")
data.2017 = read.csv("data/2017.csv")
data.2018 = read.csv("data/2018.csv")Note: this will take a relatively long time to compute as the datasets are large. Each dataset contains all shot data (including playoffs).
We’ll only look at regular season data. The playoffs in the NHL are a beast of their own.
get.regular.season = function(data) {
subset(data, isPlayoffGame == 0)
}
season.2016 = get.regular.season(data.2016)
season.2017 = get.regular.season(data.2017)
season.2018 = get.regular.season(data.2018)Now that we have our data, we can remove extraneous columns. Here is a table of what columns we are keeping, and what we are renaming them to:
| Old Column | New Column |
|---|---|
| xCordAdjusted | x |
| yCordAdjusted | y |
| goal | goal |
| shotAngleAdjusted | angle |
| goalieNameForShot | goalie_name |
| shooterName | skater_name |
| game_id | game |
Now, here is the R code to do this subsetting of the original dataset.
get.helpful.data = function(data) {
data.frame(x = data$xCordAdjusted,
y = data$yCordAdjusted,
goal = data$goal,
angle = data$shotAngleAdjusted,
goalie_name = data$goalieNameForShot,
skater_name = data$shooterName,
game = data$game_id)
}
analysis.2016 = get.helpful.data(season.2016)
analysis.2017 = get.helpful.data(season.2017)
analysis.2018 = get.helpful.data(season.2018)Now, we have all the data we need.
Function Definitions
Generic
From our data, we can calculate some very important statistics like the following:
- Goal Percent: goals per total shots
- Save Percent: saves (total shots - goals) per total shots
- Shots per Goal: total shots per goal
Additionally, we can break up our data by game. There are some generic functions we can write to help for both goalies and skaters. Let’s write them now!
get.goal.percent = function(data) {
shots = length(data$goal)
temp = subset(data, goal == 1)
goals = length(temp$goal)
goals / shots
}
get.save.percent = function(data) {
shots = length(data$goal)
temp = subset(data, goal == 1)
goals = length(temp$goal)
(shots - goals) / shots
}
get.shots.per.goal = function(data) {
shots = length(data$goal)
temp = subset(data, goal == 1)
goals = length(temp$goal)
shots / goals
}Note: when using
get.shots.per.goal, if there were no goals scored, R will handle the division by zero by returning infinity. This will be problematic when graphing data. I am still working on a good solution to this problem. Earlier, I used 200 as a substitute value. However, 200 still skews graphs, which is unideal.
get.games = function(data) {
unique(data$game)
}
get.single.game = function(data, game_id) {
subset(data, game == game_id)
}
get.all.games = function(data) {
games = get.games(data)
Map(function(x) get.single.game(data, x), games)
}Now, we can create our game by game statistic functions:
get.game.goal.percent = function(data) {
gameframe = get.all.games(data)
games.gp = map(gameframe, function(x) get.goal.percent(x))
unlist(games.gp, use.names = FALSE)
}
get.game.save.percent = function(data) {
gameframe = get.all.games(data)
games.sp = map(gameframe, function(x) get.save.percent(x))
unlist(games.sp, use.names = FALSE)
}
get.game.shots.per.goal = function(data) {
gameframe = get.all.games(data)
games.spg = map(gameframe, function(x) get.shots.per.goal(x))
unlist(games.spg, use.names = FALSE)
}Also, we’ll need a function to get match-ups between a specific goalie and skater. Let’s write that here, instead of in our goalies and our skaters sections.
get.matchup.data = function(data, goalie, skater) {
subset(data, goalie_name == goalie & skater_name == skater)
}We’ve now written our generic data handling functions.
Goalies
Let’s first start with a function to get data for a specific goalie.
get.goalie.data = function(data, name) {
subset(data, goalie_name == name)
}Skaters
Let’s first start with a function to get data for a specific skater.
get.skater.data = function(data, name) {
subset(data, skater_name == name)
}Graphing
Given specific data, we should be able to graph the location of shots. Let’s write a function that uses ggplot to do so.
graph.shot.locations = function(data, primary, secondary, name) {
plot = ggplot(data) +
geom_hex(aes(x = x, y = y, alpha = log(..count..)), fill = primary, color = secondary) +
labs(title = paste(name, "Shot Locations", sep = " "), x = "X Position", y = "Y Position") +
theme_minimal()
plot
}To see a test of what this does, let’s quickly make a graph of Roberto Luongo’s shots against him.
luongo = get.goalie.data(analysis.2017, "Roberto Luongo")
plot = graph.shot.locations(luongo, "#041E42", "#C8102E", "Roberto Luongo")
plot
Now, let’s see how it looks for a skater. Let’s look at Bryan Rust.
rust = get.skater.data(analysis.2017, "Bryan Rust")
plot = graph.shot.locations(rust, "#000000", "#FCB514", "Bryan Rust")
plot
Furthermore, we should be able to graph trends in certain statistics.
graph.trend = function(trend, type, primary, secondary, name) {
frame = data.frame(x = c(1:length(trend)), y = trend)
disp = paste(name, type, "by Game", sep = " ")
plot = ggplot(frame) +
geom_point(aes(x = x, y = y), color = primary) +
geom_smooth(aes(x = x, y = y), method = "lm", color = primary, fill = secondary) +
labs(title = disp, x = "Game Played", y = type) +
theme_minimal()
plot
}So, let’s go back to Luongo and look at his save percentage per game.
luongo.game.sp = get.game.save.percent(luongo)
plot = graph.trend(luongo.game.sp, "Save Percentage", "#041E42", "#C8102E", "Roberto Luongo")
plot
We can also see the shooter’s perspective. Let’s look at Rust’s shooting percentage (goal percent).
rust.game.gp = get.game.goal.percent(rust)
plot = graph.trend(rust.game.gp, "Shooting (Goal) Percentage", "#000000", "#FCB514", "Bryan Rust")
plot
Analysis
Goalies
Matt Murray
Matt Murray is a 24 year old phenomenon who has already won 2 Stanley Cups. Let’s take a look at how he’s done it.
murray.2016 = get.goalie.data(analysis.2016, "Matt Murray")
murray.2017 = get.goalie.data(analysis.2017, "Matt Murray")
murray.2018 = get.goalie.data(analysis.2018, "Matt Murray")Let’s start by calculating some of his stats for each season and then tabularizing.
murray.2016.sp = get.save.percent(murray.2016)
murray.2017.sp = get.save.percent(murray.2017)
murray.2018.sp = get.save.percent(murray.2018)
murray.2016.gp = get.goal.percent(murray.2016)
murray.2017.gp = get.goal.percent(murray.2017)
murray.2018.gp = get.goal.percent(murray.2018)
murray.2016.spg = get.shots.per.goal(murray.2016)
murray.2017.spg = get.shots.per.goal(murray.2017)
murray.2018.spg = get.shots.per.goal(murray.2018)Now, let’s put it in a table:
| Season | Save Percent | Goal Percent | Shots Per Goal |
|---|---|---|---|
| 2016-2017 | 0.9429599 | 0.0570401 | 17.5315315 |
| 2017-2018 | 0.9301837 | 0.0698163 | 14.3233083 |
| 2018-2019 | 0.9374185 | 0.0625815 | 15.9791667 |
We can look at his game by game data also.
murray.2016.games.sp = get.game.save.percent(murray.2016)
murray.2016.games.gp = get.game.goal.percent(murray.2016)
murray.2016.games.spg = get.game.shots.per.goal(murray.2016)
murray.2017.games.sp = get.game.save.percent(murray.2017)
murray.2017.games.gp = get.game.goal.percent(murray.2017)
murray.2017.games.spg = get.game.shots.per.goal(murray.2017)
murray.2018.games.sp = get.game.save.percent(murray.2018)
murray.2018.games.gp = get.game.goal.percent(murray.2018)
murray.2018.games.spg = get.game.shots.per.goal(murray.2018)Let’s look at his save percentage graphs:
murray.2016.sp.plot = graph.trend(murray.2016.games.sp, "Save Percentage", "#000000", "#FCB514", "Matt Murray (2016)")
murray.2016.sp.plot
murray.2017.sp.plot = graph.trend(murray.2017.games.sp, "Save Percentage", "#000000", "#FCB514", "Matt Murray (2017)")
murray.2017.sp.plot
murray.2018.sp.plot = graph.trend(murray.2018.games.sp, "Save Percentage", "#000000", "#FCB514", "Matt Murray (2018)")
murray.2018.sp.plot
And now, let’s take a look at his shot location data.
murray.2016.locations.plot = graph.shot.locations(murray.2016, "#000000", "#FCB514", "Matt Murray (2016)")
murray.2016.locations.plot
murray.2017.locations.plot = graph.shot.locations(murray.2017, "#000000", "#FCB514", "Matt Murray (2017)")
murray.2017.locations.plot
murray.2018.locations.plot = graph.shot.locations(murray.2018, "#000000", "#FCB514", "Matt Murray (2018)")
murray.2018.locations.plot
From the darkness of each hex bin, we can use that as a percent chance that Murray saved the shot. Though the darkness is not exactly the save percentage of the shots, it’s the log chance of Murray seeing the shot. This log chance is a good likelihood function of Murray saving the shot. We can see that Murray’s weakness is very close to him. Those shots are most likely to be tipped or deflected shots, not actual shots. We can see his weakness is the top of the circles, based on the X and Y coordinates.
From the graph, we can see that Murray has struggled with shots near the circles. Disregarding tips, it seems like that Murray is positionally sound with goals against near him. However, compared to the right wing, Murray seems weak against shots coming from the left wing side. If teams were to exploit this, it would be beneficial to shoot towards Murray’s glove side from the left wing circle.
Casey DeSmith
On January 11th, Casey DeSmith inked a 3 year extension with the Penguins with an average annual value of $1.25 million. His rise to earn this paycheck started with exemplary play during the 2017-2018 season and that has carried over into this season. Let’s see how he got here!
desmith.2017 = get.goalie.data(analysis.2017, "Casey DeSmith")
desmith.2018 = get.goalie.data(analysis.2018, "Casey DeSmith")Let’s start by calculating some of his stats for each season and then tabularizing.
desmith.2017.sp = get.save.percent(desmith.2017)
desmith.2018.sp = get.save.percent(desmith.2018)
desmith.2017.gp = get.goal.percent(desmith.2017)
desmith.2018.gp = get.goal.percent(desmith.2018)
desmith.2017.spg = get.shots.per.goal(desmith.2017)
desmith.2018.spg = get.shots.per.goal(desmith.2018)Now, let’s put it in a table:
| Season | Save Percent | Goal Percent | Shots Per Goal |
|---|---|---|---|
| 2017-2018 | 0.9427403 | 0.0572597 | 17.4642857 |
| 2018-2019 | 0.9422886 | 0.0577114 | 17.3275862 |
We can look at his game by game data also.
desmith.2017.games.sp = get.game.save.percent(desmith.2017)
desmith.2017.games.gp = get.game.goal.percent(desmith.2017)
desmith.2017.games.spg = get.game.shots.per.goal(desmith.2017)
desmith.2018.games.sp = get.game.save.percent(desmith.2018)
desmith.2018.games.gp = get.game.goal.percent(desmith.2018)
desmith.2018.games.spg = get.game.shots.per.goal(desmith.2018)Let’s look at his save percentage graphs:
desmith.2017.sp.plot = graph.trend(desmith.2017.games.sp, "Save Percentage", "#000000", "#FCB514", "Casey DeSmith (2017)")
desmith.2017.sp.plot
desmith.2018.sp.plot = graph.trend(desmith.2018.games.sp, "Save Percentage", "#000000", "#FCB514", "Casey DeSmith(2018)")
desmith.2018.sp.plot
And now, let’s take a look at his shot location data.
desmith.2017.locations.plot = graph.shot.locations(desmith.2017, "#000000", "#FCB514", "Casey DeSmith (2017)")
desmith.2017.locations.plot
desmith.2018.locations.plot = graph.shot.locations(desmith.2018, "#000000", "#FCB514", "Casey DeSmith (2018)")
desmith.2018.locations.plot
As we can see, DeSmith has played pretty phenomenally. He’s earned his new extension and his play has been pretty consistent. He is the perfect complement for Matt Murray.
Marc-Andre Fleury
Before he was picked by the Vegas Golden Knights in the most recent NHL Expansion Draft, Marc-Andre Fleury was the franchise goaltender for the Pittsburgh Penguins. Drafted first overall in the 2003 NHL Entry Draft, Marc-Andre “Flower” Fleury made his NHL debut in October of 2003. Since thenm he’s been the main net-protector for the Penguins (until 2017 that is). His athleticism plus is exceptionally quick reactions makes Flower a premier NHL goaltender. With 3 Stanley Cup victories (2009, 2016, and 2017), Fleury has cemented his place as a top-10 all-time goaltender. With 429 regular season wins, Fleury sits at number 9 for career regular season wins. Only two active goaltenders, Roberto Luongo and Henrik Lundqvist, sit higher above him.
Fleury has been playing phenomenally the past few season. What seems like a second-prime for the Canadian born goaltender, Fleury showed shades of his 2008-2009 season, when he led the Penguins to their third Stanley Cup
fleury.2016 = get.goalie.data(analysis.2016, "Marc-Andre Fleury")
fleury.2017 = get.goalie.data(analysis.2017, "Marc-Andre Fleury")
fleury.2018 = get.goalie.data(analysis.2018, "Marc-Andre Fleury")Let’s start by calculating some of his stats for each season and then tabularizing.
fleury.2016.sp = get.save.percent(fleury.2016)
fleury.2017.sp = get.save.percent(fleury.2017)
fleury.2018.sp = get.save.percent(fleury.2018)
fleury.2016.gp = get.goal.percent(fleury.2016)
fleury.2017.gp = get.goal.percent(fleury.2017)
fleury.2018.gp = get.goal.percent(fleury.2018)
fleury.2016.spg = get.shots.per.goal(fleury.2016)
fleury.2017.spg = get.shots.per.goal(fleury.2017)
fleury.2018.spg = get.shots.per.goal(fleury.2018)Now, let’s put it in a table:
| Season | Save Percent | Goal Percent | Shots Per Goal |
|---|---|---|---|
| 2016-2017 | 0.9314803 | 0.0685197 | 14.5943396 |
| 2017-2018 | 0.9475891 | 0.0524109 | 19.08 |
| 2018-2019 | 0.9365621 | 0.0634379 | 15.7634409 |
We can look at his game by game data also.
fleury.2016.games.sp = get.game.save.percent(fleury.2016)
fleury.2016.games.gp = get.game.goal.percent(fleury.2016)
fleury.2016.games.spg = get.game.shots.per.goal(fleury.2016)
fleury.2017.games.sp = get.game.save.percent(fleury.2017)
fleury.2017.games.gp = get.game.goal.percent(fleury.2017)
fleury.2017.games.spg = get.game.shots.per.goal(fleury.2017)
fleury.2018.games.sp = get.game.save.percent(fleury.2018)
fleury.2018.games.gp = get.game.goal.percent(fleury.2018)
fleury.2018.games.spg = get.game.shots.per.goal(fleury.2018)Let’s look at his save percentage graphs:
fleury.2016.sp.plot = graph.trend(fleury.2016.games.sp, "Save Percentage", "#000000", "#FCB514", "Marc-Andre Fleury (2016)")
fleury.2016.sp.plot
fleury.2017.sp.plot = graph.trend(fleury.2017.games.sp, "Save Percentage", "#B4975A", "#333F42", "Marc-Andre Fleury (2017)")
fleury.2017.sp.plot
fleury.2018.sp.plot = graph.trend(fleury.2018.games.sp, "Save Percentage", "#B4975A", "#333F42", "Marc-Andre Fleury (2018)")
fleury.2018.sp.plot
And now, let’s take a look at his shot location data.
fleury.2016.locations.plot = graph.shot.locations(fleury.2016, "#000000", "#FCB514", "Marc-Andre Fleury (2016)")
fleury.2016.locations.plot
fleury.2017.locations.plot = graph.shot.locations(fleury.2017, "#B4975A", "#333F42", "Marc-Andre Fleury (2017)")
fleury.2017.locations.plot
fleury.2018.locations.plot = graph.shot.locations(fleury.2018, "#B4975A", "#333F42", "Marc-Andre Fleury (2018)")
fleury.2018.locations.plot
Fleury has shown exceptional consistency. For any young goaltender breaking into the league, he should be a top-role model.